Question-answer generation system, method, and program

Abstract

To flexibly generate a question-answer pair from interaction information.SOLUTION: According to one embodiment, a question-answer generation system includes an acquisition unit configured to acquire one or more pieces of dialogue information, a summary generation unit configured to generate one or more summaries from each piece of dialogue information using a summary generation model, a question generation unit configured to generate one or more questions from a plurality of summaries using the question generation model, a recommendation unit configured to recommend registration of an answer to the one or more questions, and a storage unit configured to store question-answer information indicating a question-answer pair that is a set of a registered question and answer.SELECTED DRAWING: Figure 11

Description

[Technical Field]

[0001] The present invention relates to a system, method, and program for generating questions and answers. [Background technology]

[0002] In recent years, automated response services using AI (Artificial Intelligence) have become widespread. In automated response services, pairs of questions and answers (hereinafter referred to as "question-answer pairs") are registered in advance, and the AI ​​identifies a registered question that corresponds to a question from a questioner, and then provides the questioner with a registered answer that corresponds to the identified question.

[0003] As a method for registering pairs of questions and answers, Cited Document 1 discloses a question and answer collection generation system including a question and answer information storage unit in which response pairs each combining a question and an answer are recorded, and a control unit that generates a question and answer collection, in which the control unit classifies utterances in one session into user utterances and operator utterances according to the source of the message, and for each operator utterance in the session, creates a plurality of response pairs each configured with the operator utterance as an answer and the user utterance and operator utterance preceding the operator utterance as a question, calculates features of each response pair, compares them with features of registered response pairs, and excludes the pair as a candidate for registration if the similarity is higher than a threshold, displays the response pairs identified using the features, and records the response pairs confirmed by an administrator in the question and answer collection. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7143460 Summary of the Invention [Problem to be solved by the invention]

[0005] In the above-mentioned conventional technology, the utterances contained in a message had to be distinguished into user utterances and operator utterances depending on the sender, and therefore question-and-answer pairs (response pairs in Cited Document 1) could only be generated from information that clearly identified the sender of each utterance.

[0006] The present invention has been made in view of the above-mentioned problems, and has an object to enable question and answer pairs to be flexibly generated from dialogue information. [Means for solving the problem]

[0007] According to one embodiment, the question and answer generation system includes an acquisition unit that acquires one or more pieces of dialogue information, a summary generation unit that uses a summary generation model to generate one or more summaries from each piece of dialogue information, a question generation unit that uses a question generation model to generate one or more questions from the multiple summaries, a recommendation unit that recommends registering answers to the one or more questions, and a memory unit that stores question and answer information that indicates question and answer pairs, which are combinations of registered questions and answers. [Effects of the Invention]

[0008] According to one embodiment, question and answer pairs can be flexibly generated from dialogue information. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 illustrates an example of the configuration of a question and answer generation system 1000. [Figure 2] 1 is a diagram illustrating an example of a hardware configuration of an information processing device 100. FIG. [Figure 3] 1 is a diagram illustrating an example of a functional configuration of a question and answer generation device 1. FIG. [Figure 4] FIG. 2 is a diagram illustrating an example of the functional configuration of a user terminal 2. [Figure 5] 1 is a flowchart showing an example of processing executed by the question and answer generation system 1000. [Figure 6] FIG. 10 is a diagram showing an example of dialogue information 121. [Figure 7] FIG. 10 is a diagram showing an example of summary information 123. [Figure 8] FIG. 10 is a diagram illustrating an example of a clustering result. [Figure 9] FIG. 10 is a diagram illustrating an example of a clustering result. [Figure 10] FIG. 2 is a diagram showing an example of answer to question information 222. [Figure 11] FIG. 10 is a diagram showing an example of a registration screen sc. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, each embodiment of the present invention will be described with reference to the accompanying drawings. Note that, in the description of the specification and drawings relating to each embodiment, components having substantially the same functional configuration are designated by the same reference numerals, and redundant description will be omitted.

[0011] <System configuration> First, an overview of the question and answer generation system 1000 according to this embodiment will be described. The question and answer generation system 1000 is an information processing system that supports the generation of question and answer pairs by automatically generating questions based on dialogue information and recommending to users the registration of answers to the generated questions.

[0012] The question and answer pairs generated by the question and answer generation system 1000 may be posted on a website as FAQs (Frequently Asked Questions), may be used in a chatbot that answers questions from questioners via chat, or may be used in an interactive voice response (IVR) that answers questions from questioners over the phone.

[0013] Fig. 1 is a diagram showing an example of the configuration of a question and answer generation system 1000. As shown in Fig. 1, the question and answer generation system 1000 includes a question and answer generation device 1 and a user terminal 2, which are communicably connected to each other via a network N. The network N is, for example, a wired LAN (Local Area Network), a wireless LAN, the Internet, a public line network, a mobile data communication network, or a combination of these. In the example of Fig. 1, the question and answer generation system 1000 includes one question and answer generation device 1 and one user terminal 2, but may include multiple of each.

[0014] The question and answer generation device 1 is an information processing device that automatically generates questions based on dialogue information and recommends to users that they register answers to the generated questions. The question and answer generation device 1 is, for example, but is not limited to, a PC (Personal Computer), a smartphone, a tablet terminal, a server device, or a microcomputer. In the example of FIG. 1, the question and answer generation device 1 is a single information processing device, but it may also be realized as a system consisting of multiple information processing devices connected via a network N.

[0015] The user terminal 2 is an information processing device used by a user. A user is a person who uses the question and answer generation system 1000, and uses the user terminal 2 to register answers to questions generated by the question and answer generation system 1000. The user terminal 2 is, for example, but is not limited to, a PC, a smartphone, or a tablet terminal.

[0016] <Hardware configuration of information processing device 100> Next, a description will be given of the hardware configuration of the information processing device 100. Fig. 2 is a diagram showing an example of the hardware configuration of the information processing device 100. As shown in Fig. 2, the information processing device 100 includes a processor 101, a memory 102, a storage 103, a communication I / F 104, an input device 105, an output device 106, and a drive device 107, which are connected to each other via a bus B.

[0017] The processor 101 controls each component of the information processing device 100 and realizes the functions of the information processing device 100 by loading various programs including an OS (Operating System) stored in the storage 103 into the memory 102 and executing the programs. The processor 101 is, for example, but not limited to, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), or a DSP (Digital Signal Processor).

[0018] The memory 102 is, for example, a read-only memory (ROM), a random access memory (RAM), or a combination thereof. The ROM is, for example, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a combination thereof. The RAM is, for example, but not limited to, a dynamic random access memory (DRAM) or a static random access memory (SRAM).

[0019] The storage 103 stores various programs including an OS and data. The storage 103 is, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or storage class memories (SCM), but is not limited to these.

[0020] The communication I / F 104 is an interface for connecting the information processing device 100 to an external device via the network N and controlling communication. The communication I / F 104 is, for example, Bluetooth (registered trademark), Wi-Fi (registered trademark), ZigBee (registered trademark), or Ethernet (registered trademark), but is not limited to these.

[0021] The input device 105 is a device for inputting information to the information processing device 100. The input device 105 is, for example, a mouse, a keyboard, a touch panel, a microphone, a scanner, an imaging device (camera), various sensors, or an operation button, but is not limited to these.

[0022] The output device 106 is a device for outputting information from the information processing device 100. The output device 106 is, for example, a display device, a projector, a printer, a speaker, or a vibrator, but is not limited to these.

[0023] The drive device 107 is a device that reads and writes data from and to the recording medium 108. The drive device 107 is, for example, but not limited to, a magnetic disk drive, an optical disk drive, a magneto-optical disk drive, or an SD card reader. The recording medium 108 is, for example, but not limited to, a CD (Compact Disc), a DVD (Digital Versatile Disc), an FD (Floppy Disk), an MO (Magneto-Optical disk), a BD (Blu-ray (registered trademark) Disc), a USB (registered trademark) memory, or an SD card.

[0024] In this embodiment, the program may be written to the memory 102 or the storage 103 during the manufacturing stage of the information processing device 100, or may be provided to the information processing device 100 via the network N, or may be provided to the information processing device 100 via a non-transitory computer-readable recording medium such as the recording medium 108.

[0025] <Functional Configuration of Question and Answer Generation Device 1> Next, we will explain the functional configuration of the question and answer generation device 1. Fig. 3 is a diagram showing an example of the functional configuration of the question and answer generation device 1. As shown in Fig. 3, the question and answer generation device 1 includes a communication unit 11, a storage unit 12, and a control unit 13.

[0026] The communication unit 11 is realized by the communication I / F 104. The communication unit 11 transmits and receives information to and from the user terminal 2 via the network N.

[0027] The storage unit 12 is realized by the memory 102 and the storage 103. The storage unit 12 stores dialogue information 121, a summary generation model 122, summary information 123, a question generation model 124, an answer candidate generation model 125, and question answer information 126.

[0028] The dialogue information 121 is information indicating a dialogue between a questioner and a respondent, and includes information indicating a question from the questioner and information indicating an answer from the respondent. One piece of dialogue information includes one or more pairs of questions and answers. The dialogue information may be audio information of a telephone conversation between the questioner and the respondent, or text information of a chat between the questioner and the respondent. The respondent may be a person or a system (chatbot or automated voice response). For example, the questioner may be a user of a service and the respondent may be a provider of the service, but this is not limited to this. The memory unit 12 stores one or more pieces of dialogue information 121 for each respondent.

[0029] The summary generation model 122 is a trained machine learning model that has been trained to be able to output a summary of the dialogue information 121 when the dialogue information 121 is input. The summary generation model 122 is preferably a generation model that generates text in accordance with a prompt. The summary generation model 122 is, for example, a large language model (LLM) or a large multimodal model, but is not limited to these. The summary generation model 122 may be stored in the storage unit 12, or may be stored in an external server communicatively connected to the question and answer generation device 1 via the network N and used via an API (Application Programming Interface).

[0030] The summary information 123 is information that summarizes the content of the question from the questioner that is included in the dialogue information 121. The summary information 123 is, for example, information that indicates a sentence that briefly expresses the content of the question from the questioner that is included in the dialogue information 121, but is not limited to this. One or more pieces of summary information 123 are stored for one piece of dialogue information 121.

[0031] The question generation model 124 is a trained machine learning model that has been trained to be able to output question information when summary information 123 is input. The question generation model 124 is preferably a generation model that generates text in response to a prompt. The question information is information indicating a question, and corresponds to information that summarizes one or more pieces of summary information 123. The question referred to here may be a sentence (question sentence) expressing the content of the question, or a title expressing the content of the question. The question generation model 124 is, for example, but is not limited to, a large-scale language model or a large-scale multimodal model. The question generation model 124 may be stored in the storage unit 12, or may be stored in an external server communicatively connected to the question and answer generation device 1 via the network N and used by an API. Furthermore, the question generation model 124 may be the same machine learning model as the summary generation model 122, or may be a different machine learning model.

[0032] The answer candidate generation model 125 is a trained machine learning model that has been trained to be able to output answer candidate information when question information is input. The answer candidate generation model 125 is preferably a generation model that generates text in response to a prompt. The answer candidate information is information indicating answer candidates to a question. The answer candidate generation model 125 generates answer candidate information based on reference information. The reference information includes dialogue information 121 corresponding to the question, public information published online (e.g., the answerer's website), and information uploaded by the user via the user terminal 2 (e.g., text information such as a manual). The answer candidate generation model 125 is, for example, but is not limited to, a large-scale language model or a large-scale multimodal model. The answer candidate generation model 125 may be stored in the storage unit 12, or may be stored in an external server communicatively connected to the question and answer generation device 1 via the network N and used via an API. The answer candidate generation model 125 may be the same machine learning model as at least one of the summary generation model 122 and the question generation model 124, or may be a different machine learning model.

[0033] The question answer information 126 is information indicating a question and answer pair, which is a combination of a question and an answer. The question answer information 126 includes question information, answer information, and answer candidate information. The answer information is information indicating an answer to a question, and is registered by a user. The storage unit 12 stores one or more pieces of question answer information 126 for each answerer.

[0034] The control unit 13 is realized by the processor 101 reading and executing a program from the memory 102 and working in cooperation with other hardware components. The control unit 13 controls the overall operation of the question and answer generation device 1. The control unit 13 includes an acquisition unit 131, a summary generation unit 132, a question generation unit 133, and an answer candidate generation unit 134.

[0035] The acquisition unit 131 acquires one or more pieces of dialogue information 121 and stores them in the storage unit 12.

[0036] The summary generation unit 132 uses the summary generation model 122 to generate one or more summaries from each piece of dialogue information 121, and stores summary information 123 indicating the generated summaries in the storage unit 12.

[0037] The question generation unit 133 uses the question generation model 124 to generate one or more questions from the multiple summaries, and stores question information indicating the generated questions as answer-to-question information 126 in the storage unit 12.

[0038] The answer candidate generation unit 134 uses the answer candidate generation model 125 to generate answer candidates for the question based on the reference information, and stores answer candidate information indicating the generated answer candidates in the storage unit 12 as question answer information 126.

[0039] The functional configuration of the question and answer generation device 1 is not limited to the above example. For example, the question and answer generation device 1 may have some of the above functional configuration, with the rest being provided in the user terminal 2. The question and answer generation device 1 may also have functional configurations other than those described above. Furthermore, each functional configuration of the question and answer generation device 1 may be realized by software, as described above, or by hardware such as an IC chip, a SoC (System on Chip), an LSI (Large Scale Integration), a microcomputer, etc.

[0040] <Functional configuration of user terminal 2> Next, a description will be given of the functional configuration of the user terminal 2. Fig. 4 is a diagram showing an example of the functional configuration of the user terminal 2. As shown in Fig. 3, the user terminal 2 includes a communication unit 21, a storage unit 22, and a control unit 23.

[0041] The communication unit 21 is realized by the communication I / F 204. The communication unit 21 transmits and receives information to and from the question and answer generation device 1 via the network N.

[0042] The storage unit 22 is realized by the memory 202 and the storage 203. The storage unit 22 stores dialogue information 221 and question and answer information 222.

[0043] The dialogue information 221 is information indicating a dialogue between a questioner and a respondent (user), and includes information indicating a question from the questioner and information indicating an answer from the respondent. One piece of dialogue information includes one or more pairs of questions and answers. The dialogue information may be audio information of a telephone conversation between the questioner and the respondent, or text information of a chat between the questioner and the respondent. The respondent may be a person (user) or a system (chatbot or automated voice response) operated by the user. For example, the questioner may be a user of a service and the respondent may be a provider of the service, but this is not limited to this. The storage unit 22 stores one or more pieces of dialogue information 221 of a user.

[0044] The question answer information 222 is information indicating a question and answer pair, which is a combination of a question and an answer. The question answer information 222 includes question information, answer information, and answer candidate information. The answer information is information indicating an answer to a question, and is registered by a user. The storage unit 22 stores the question answer information 222 of the user.

[0045] The control unit 23 is realized by the processor 201 reading and executing a program from the memory 202 and working in cooperation with other hardware configurations. The control unit 23 controls the overall operation of the user terminal 2. The control unit 23 includes an acquisition unit 231 and a recommendation unit 232.

[0046] The acquiring unit 231 acquires the answer to question information 126 of the user from the question and answer generation device 1 and stores it in the storage unit 22 as answer to question information 222.

[0047] The recommendation unit 232 recommends to the user to register answers to one or more questions.

[0048] The functional configuration of the user terminal 2 is not limited to the above example. For example, the user terminal 2 may have some of the above functional configurations and the user terminal 2 may have the rest. The user terminal 2 may also have functional configurations other than those described above. Each functional configuration of the user terminal 2 may be realized by software, as described above, or by hardware such as an IC chip, SoC, LSI, or microcomputer.

[0049] <Processing Executed by Question and Answer Generation System 1000> Next, a description will be given of the processing executed by the question and answer generation system 1000. Figure 5 is a flowchart showing an example of the processing executed by the question and answer generation system 1000.

[0050] (Step S101) The user terminal 2 transmits one or more pieces of dialogue information 221, which are stored in the storage unit 22 and indicate a dialogue between the questioner and the user (answerer), to the question and answer generation device 1. The one or more pieces of dialogue information 221 are text information, audio information, or a combination thereof.

[0051] (Step S102) The acquisition unit 131 of the question and answer generation device 1 acquires the dialogue information 221 from the user terminal 2 and stores it in the storage unit 12 as dialogue information 121 .

[0052] Fig. 6 is a diagram showing an example of the dialogue information 121. In the example of Fig. 6, the user is the xxx restaurant, and the questioner is a customer of the xxx restaurant. Fig. 6 shows three pieces of dialogue information 121A to 121C, and the dialogue information 121A is information indicating a phone call from the customer (questioner) to the user (answerer) inquiring about a reservation.

[0053] (Step S103) The summary generation unit 132 uses the summary generation model 122 stored in the storage unit 12 to generate one or more summaries from each piece of dialogue information 121 stored in the storage unit 12. Specifically, the summary generation unit 132 inputs a prompt to the summary generation model 122 instructing it to summarize the question from the questioner, which is included in the dialogue information 121, and obtains summary information 123 indicating the summary output by the summary generation model 122. The summary generation unit 132 saves the obtained summary information 123 in the storage unit 12.

[0054] Fig. 7 is a diagram showing an example of summary information 123. In the example of Fig. 7, summaries "inquiry about reservation," "confirmation of business days," and "confirmation of menu" are generated from dialogue information 121A, a summary "question about business hours" is generated from dialogue information 121B, and a summary "inquiry about business hours," "confirmation of ingredients with allergies," and "directions to the store" are generated from dialogue information 121C. As shown in Fig. 7, even summaries for questions about the same content (business hours) may not be the same.

[0055] If the dialogue information 121 is text information, the summary generation unit 132 may generate a summary from the text information. If the dialogue information 121 is audio information, the summary generation unit 132 may convert the audio information into text information before generating a summary, or may generate a summary from the audio information itself.

[0056] (Step S104) The question generation unit 133 uses the question generation model 124 stored in the storage unit 12 to generate one or more questions from the multiple summaries stored in the storage unit 12. The question generation unit 133 stores question information indicating the generated question in the storage unit 12 as question answer information 126. A specific example of a method for generating a question will be described below.

[0057] ·Question generation method 1 The question generation unit 133 inputs a prompt to the question generation model 124 to instruct it to generate a question from multiple pieces of summary information 123 stored in the storage unit 12, and acquires question information indicating the question output by the question generation model 124. The question generation unit 133 may input a prompt to the question generation model 124 to instruct it to generate a frequently occurring question. This makes it possible to exclude rare questions and generate frequently occurring questions preferentially. In the example of FIG. 7, seven summaries are input to the question generation model 124, and one or more questions are generated.

[0058] Next, the question generation unit 133 refers to the question-and-answer information 126 and calculates the similarity between each question registered as a question-and-answer pair and each question generated using the question generation model 124. For example, the question generation unit 133 generates a distributed representation of each question and calculates the cosine similarity between the distributed representation of the registered question and the distributed representation of the generated question. Note that the method for calculating the similarity is arbitrary.

[0059] Then, the question generation unit 133 deletes, from among the generated questions, questions whose similarity to already registered questions is equal to or greater than a threshold, and stores question information indicating questions whose similarity is less than the threshold as question answer information 126 in the storage unit 12. This prevents the registration of answers to questions similar to already registered questions from being recommended, thereby preventing duplicate registration of question-answer pairs.

[0060] The question generation unit 133 may input a prompt to the question generation model 124 to instruct it to determine whether each question registered as a question-and-answer pair is similar to each question generated using the question generation model 124, exclude questions determined to be similar to the registered questions, and store question information indicating questions determined to be dissimilar in the memory unit 12 as question-and-answer information 126.

[0061] ·Question generation method 2 The question generation unit 133 clusters the summaries indicated by the multiple pieces of summary information 123 stored in the storage unit 12. For example, the question generation unit 133 generates a distributed representation of each summary and clusters the summaries using the k-nearest neighbor method. Note that any clustering method may be used.

[0062] Fig. 8 is a diagram showing an example of the clustering result. In the example of Fig. 8, four clusters C1 to C4 are generated from seven summaries. As shown in Fig. 8, summaries with similar content are classified into the same cluster.

[0063] Next, the question generation unit 133 inputs a prompt to the question generation model 124 to instruct it to generate a question from the abstract included in each cluster, and obtains question information indicating the question output by the question generation model 124. In this way, a question is generated for each cluster.

[0064] In the example of FIG. 8, four questions corresponding to clusters C1 to C4 are generated. The question generation unit 133 may input a prompt to the question generation model 124 to instruct it to generate questions only for clusters that include a predetermined number of summaries. This allows rare questions to be excluded and frequently occurring questions to be generated preferentially. For example, in the example of FIG. 8, when questions are generated only for clusters that include two or more summaries, two questions corresponding to clusters C1 and C2 are generated.

[0065] Next, the question generation unit 133 refers to the question and answer information 126 and calculates the similarity between each question registered as a question and answer pair and each question generated using the question generation model 124. The method for calculating the similarity is as described above.

[0066] Then, the question generation unit 133 deletes, from among the generated questions, questions whose similarity to already registered questions is equal to or greater than a threshold, and stores question information indicating questions whose similarity is less than the threshold as question answer information 126 in the storage unit 12. This prevents the registration of answers to questions similar to already registered questions from being recommended, thereby preventing duplicate registration of question-answer pairs.

[0067] The question generation unit 133 may input a prompt to the question generation model 124 to instruct it to determine whether each question registered as a question-and-answer pair is similar to each question generated using the question generation model 124, exclude questions determined to be similar to the registered questions, and store question information indicating questions determined to be dissimilar in the memory unit 12 as question-and-answer information 126.

[0068] ·Question generation method 3 The question generation unit 133 clusters the summaries indicated by the plurality of pieces of summary information 123 stored in the storage unit 12 and the registered questions (questions included in question-and-answer pairs) indicated by the question information included in the question-and-answer information 126. For example, the question generation unit 133 generates a distributed representation of each summary and each question, and clusters each summary and each question using the k-nearest neighbor method. Note that any clustering method may be used.

[0069] Fig. 9 is a diagram showing an example of the clustering results. In the example of Fig. 9, four clusters C1 to C4 are generated from seven summaries and two questions. Cluster C2 includes a registered question called "Menu," and cluster C3 includes a registered question called "How to make a reservation." As shown in Fig. 9, summaries and questions with similar content are classified into the same cluster.

[0070] Next, the question generation unit 133 inputs a prompt to the question generation model 124 to instruct it to generate questions from summaries included in each of the generated clusters that does not include a registered question, and obtains question information indicating the questions output by the question generation model 124. As a result, questions are generated for each cluster from the summaries of the clusters that do not include a registered question.

[0071] In the example shown in Figure 9, two questions corresponding to clusters C1 and C2 are generated. This prevents the recommendation of registering answers to questions similar to questions that have already been registered, thereby preventing duplicate registration of question-answer pairs.

[0072] The question generation unit 133 may input a prompt to the question generation model 124 to instruct it to generate questions only for clusters that contain a predetermined number of summaries. This allows rare questions to be excluded and frequently occurring questions to be generated preferentially. For example, when questions are generated only for clusters that contain two or more summaries, in the example of FIG. 9, one question corresponding to cluster C1 is generated.

[0073] (Step S105) The answer candidate generation unit 134 uses the answer candidate generation model 125 stored in the storage unit 12 to generate answer candidates for the question generated in step S104 based on the reference information. Specifically, when the question is generated by question generation method 1, the answer candidate generation unit 134 inputs a prompt to the summary generation model 122 to instruct it to generate answer candidates from all of the dialogue information 121, and acquires answer candidate information indicating the answer candidates output by the answer candidate generation model 125. When the question is generated by question generation method 2 or 3, the answer candidate generation unit 134 inputs a prompt to the summary generation model 122 to instruct it to generate answer candidates from the dialogue information 121 corresponding to the question, and acquires answer candidate information indicating the answer candidates output by the answer candidate generation model 125. The dialogue information 121 corresponding to the question is dialogue information 121 whose summary is included in the cluster from which the question was generated. The answer candidate generation unit 134 may also input a prompt to the summary generation model 122 instructing the summary generation model 122 to generate answer candidates from public information published online (for example, the respondent's website), and acquire answer candidate information indicating the answer candidates output by the answer candidate generation model 125. The answer candidate generation unit 134 may also input a prompt to the summary generation model 122 instructing the summary generation model 122 to generate answer candidates from information uploaded by the user (for example, text information such as a manual), and acquire answer candidate information indicating the answer candidates output by the answer candidate generation model 125. The answer candidate generation unit 134 saves the answer candidate information indicating the generated answer candidates in the storage unit 12 as question answer information 126.

[0074] (Step S106) Thereafter, the question and answer generation device 1 transmits the answer to question information 126 stored in the storage unit 12 to the user terminal 2.

[0075] Fig. 10 is a diagram showing an example of the answer-to-question information 126. The answer-to-question information 126 in Fig. 10 includes information items such as "PID," "question," "answer," and "candidate answer."

[0076] "PID" is identification information that uniquely identifies a question-and-answer pair. "Question" is information indicating a question generated by the question generation unit 133. In the example of FIG. 10, a title indicating the content of the question is generated as the question. "Answer" is information indicating an answer registered by the user. "Answer candidate" is information indicating an answer candidate generated by the answer candidate generation unit 134.

[0077] Question and answer pairs P01 and P02 are question and answer pairs in which answers to the questions have already been registered. Question and answer pairs P03 and P04 are question and answer pairs in which answers to the questions have not yet been registered. The questions in question and answer pairs P03 and P04 are questions newly generated in steps S101 to S105, and answer candidates have been registered instead of answers.

[0078] (Step S107) The acquisition unit 231 of the user terminal 2 acquires the answer to question information 126 received from the question and answer generation device 1 and stores it in the storage unit 22 as answer to question information 222.

[0079] (Step S108) The recommendation unit 232 displays the question and answer information 222 on a display device and recommends to the user that an answer be registered for a question to which no answer has been registered.

[0080] Fig. 11 is a diagram showing an example of a registration screen sc. The registration screen sc is a screen for registering answers to questions. The registration screen sc in Fig. 11 includes a flag sc1, text boxes sc2 to sc4, and buttons sc5 to sc9.

[0081] The flag sc1 indicates the registration status of the answer to the question (registration information of the question-and-answer pair). The flag sc1 is displayed for each question. In the example of FIG. 11, the flag sc1 indicating unregistered is highlighted in a different color from the flag sc1 indicating registered.

[0082] The text box sc2 is a text box in which the question indicated by the question information included in the answer to question information 222 is displayed. In the example of FIG. 11, a title indicating the content of the question is displayed as the question. A text box sc2 is displayed for each question. In the example of FIG. 11, four text boxes sc2 are displayed, displaying two questions for which answers have been registered and two questions for which answers generated by the question and answer generation device 1 have not been registered. The user can input and modify questions from the text boxes sc2.

[0083] The text box sc3 is a text box in which the answer indicated by the answer information included in the question and answer information 222 is displayed. A text box sc3 is displayed for each question. In the example of FIG. 11, the text box sc3 corresponding to a question for which an answer has not been registered is blank, while the text box sc3 corresponding to a question for which an answer has been registered displays the registered answer. The user can input and modify a question from the text box sc3.

[0084] The text box sc4 is a text box in which answer candidates indicated by the answer candidate information included in the answer to question information 222 are displayed. The text box sc4 is displayed for each question for which an answer has not been registered. In the example of FIG. 11, the text box sc4 corresponding to a question for which an answer has not been registered displays the answer candidates corresponding to the question. In this way, displaying answer candidates for a question can assist the user in entering an answer to the question.

[0085] Button sc5 is a button for inputting the answer candidate displayed in text box sc4 into text box sc3. Button sc5 is displayed for each text box sc4. When the user selects button sc5, the answer candidate displayed in text box sc4 corresponding to button sc5 is input into the corresponding text box sc3.

[0086] Button sc6 is a button for deleting a registered question and answer pair. Button sc6 is displayed for each question for which an answer has been registered. When the user selects button sc6, the user terminal 2 deletes the question and answer information 222 of the corresponding question and answer pair.

[0087] The button sc7 is a button for returning to the screen before the registration screen sc. When the user selects the button sc7, the user terminal 2 displays the screen before the registration screen sc on the display device.

[0088] Button sc8 is a button for registering a question and answer pair. When the user selects button sc8, the user terminal 2 stores question and answer information 222, which includes question information indicating the question entered in text box sc2 and answer information indicating the answer entered in text box sc3, in the storage unit 22, and transmits it to the question and answering apparatus 1.

[0089] Button sc9 is a button for adding a question. When the user selects button sc9, the user terminal 2 displays a flag sc1 indicating that a question has not been registered, and blank text boxes sc2 and sc3. The user can register a desired question-and-answer pair by entering the question and answer in the displayed text boxes sc2 and sc3.

[0090] In this way, by displaying the question answer information 222 on the registration screen sc, it is possible to recommend to the user that they register an answer to a question for which an answer has not been registered. Note that the registration screen sc is not limited to the example in Fig. 11. The registration screen sc may be any screen that allows the registration of an answer to a question for which an answer has not been registered.

[0091] (Step S109) When the registration screen sc is displayed, the user inputs an answer to a question for which an answer has not been registered and requests registration of the input answer. In the example of FIG. 11, the user inputs an answer in text box sc3 and selects button sc8. The user terminal 2 accepts the input of the answer and, when requested to register, saves answer information indicating the accepted answer as question answer information 222 in the storage unit 22. At this time, the user may also modify the recommended question, modify a registered question-and-answer pair, or add a new question-and-answer pair from the registration screen sc. The question information indicating the revised question and the answer information indicating the revised answer are saved in the storage unit 22 as question answer information 222.

[0092] (Step S110) The user terminal 2 transmits the answer to question information 222 stored in the storage unit 22 to the question and answer generation device 1.

[0093] (Step S111) When the acquisition unit 131 of the question and answer generation device 1 acquires the question and answer information 222 from the user terminal 2, it stores the information in the storage unit 12 as question and answer information 126. As a result, an answer (a new question and answer pair) corresponding to the recommended question is registered.

[0094] <Summary> As described above, according to this embodiment, a question and answer generation system 1000 is realized, which includes an acquisition unit 131 that acquires one or more pieces of dialogue information 121, a summary generation unit 132 that uses a summary generation model 122 to generate one or more summaries from each piece of dialogue information 121, a question generation unit 133 that uses a question generation model 124 to generate one or more questions from the multiple summaries, a recommendation unit 232 that recommends registering answers to one or more questions, and a memory unit 12 that stores question and answer information 126 that indicates question and answer pairs, which are combinations of registered questions and answers.

[0095] The question and answer generation system 1000 generates questions from dialogue information 121, recommends to the user the registration of answers to the generated questions, and can support the registration of question and answer pairs. The dialogue information 121 used to generate questions can be text information or audio information, and additional information such as the sender is not required. Therefore, the question and answer generation system 1000 can flexibly generate question and answer pairs from any dialogue information 121.

[0096] <Additional Notes> The present embodiment includes the following disclosure.

[0097] (Appendix 1) an acquisition unit that acquires one or more pieces of dialogue information; a summary generation unit that generates one or more summaries from each piece of dialogue information using a summary generation model; a question generation unit that generates one or more questions from a plurality of summaries using a question generation model; a recommendation unit that recommends the registration of an answer to the one or more questions; a storage unit that stores question and answer information indicating question and answer pairs that are combinations of registered questions and answers; A question and answer generation system comprising:

[0098] (Appendix 2) The question generator clusters the plurality of summaries and generates the question for each cluster. 10. The question and answer generation system according to claim 1.

[0099] (Appendix 3) The question generation unit clusters the plurality of summaries and the questions included in the question-answer pairs, and generates the questions for clusters that do not include the questions included in the question-answer pairs. 10. The question and answer generation system according to claim 1.

[0100] (Appendix 4) The question generator excludes, from the one or more questions, a question whose similarity to a question included in the question-and-answer pair is equal to or greater than a threshold. 10. The question and answer generation system according to claim 1.

[0101] (Appendix 5) The question generation unit excludes a question determined by the question generation model to be similar to a question included in the question and answer pair from the one or more questions. 10. The question and answer generation system according to claim 1.

[0102] (Appendix 6) an answer candidate generation unit that generates answer candidates for the question based on the reference information by utilizing the answer candidate generation model; The recommendation unit recommends the answer candidates. 10. The question and answer generation system according to claim 1.

[0103] (Appendix 7) The reference information includes the dialogue information corresponding to the question. 7. The question and answer generation system according to claim 6.

[0104] (Appendix 8) The reference information includes public information published online. 7. The question and answer generation system according to claim 6.

[0105] (Appendix 9) The reference information includes information uploaded by the user. 7. The question and answer generation system according to claim 6.

[0106] (Appendix 10) The dialogue information is text information or audio information. 10. The question and answer generation system according to claim 1.

[0107] (Appendix 11) At least one of the summary generation model and the question generation model is a large-scale language model or a large-scale multimodal model. 10. The question and answer generation system according to claim 1.

[0108] (Appendix 12) A question generation method executed by a question and answer generation system, comprising: acquiring one or more pieces of dialogue information; generating one or more summaries from each of the dialogue information using a summary generation model; generating one or more questions from the plurality of summaries using a question generation model; recommending the registration of an answer to the one or more questions; storing question and answer information indicating question and answer pairs, which are combinations of registered questions and answers; A question generation method including:

[0109] (Appendix 13) For the question and answer generation system, acquiring one or more pieces of dialogue information; generating one or more summaries from each of the dialogue information using a summary generation model; generating one or more questions from the plurality of summaries using a question generation model; recommending the registration of an answer to the one or more questions; storing question and answer information indicating question and answer pairs, which are combinations of registered questions and answers; A program for executing a question generation method including the steps of:

[0110] The embodiments disclosed herein are illustrative in all respects and should not be considered limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims. Furthermore, the present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0111] 1: Question and answer generator 2: User terminal 121: Dialogue information 122: Summary generation model 123: Summary information 124: Question generation model 125: Answer candidate generation model 126:Question and answer information 131: Acquisition Department 132:Summary generator 133: Question generation section 134:Answer candidate generator

Claims

1. an acquisition unit that acquires one or more pieces of dialogue information; a summary generation unit that generates one or more summaries from each piece of dialogue information using a summary generation model; a question generation unit that generates one or more questions from a plurality of summaries using a question generation model; a recommendation unit that recommends the registration of an answer to the one or more questions; a storage unit that stores question and answer information indicating question and answer pairs that are combinations of registered questions and answers; A question and answer generation system comprising:

2. The question generator clusters the plurality of summaries and generates the question for each cluster. The question and answer generation system according to claim 1 .

3. The question generation unit clusters the plurality of summaries and the questions included in the question-answer pairs, and generates the questions for clusters that do not include the questions included in the question-answer pairs. The question and answer generation system according to claim 1 .

4. The question generator excludes, from the one or more questions, a question whose similarity to a question included in the question-and-answer pair is equal to or greater than a threshold. The question and answer generation system according to claim 1 .

5. The question generation unit excludes a question determined by the question generation model to be similar to a question included in the question and answer pair from the one or more questions. The question and answer generation system according to claim 1 .

6. an answer candidate generation unit that generates answer candidates for the question based on the reference information by utilizing the answer candidate generation model; The recommendation unit recommends the answer candidates. The question and answer generation system according to claim 1 .

7. The reference information includes the dialogue information corresponding to the question. The question and answer generation system according to claim 6 .

8. The reference information includes public information published online. The question and answer generation system according to claim 6 .

9. The reference information includes information uploaded by the user. The question and answer generation system according to claim 6 .

10. The dialogue information is text information or audio information. The question and answer generation system according to claim 1 .

11. At least one of the summary generation model and the question generation model is a large-scale language model or a large-scale multimodal model. The question and answer generation system according to claim 1 .

12. A question generation method executed by a question and answer generation system, comprising: acquiring one or more pieces of dialogue information; generating one or more summaries from each of the dialogue information using a summary generation model; generating one or more questions from the plurality of summaries using a question generation model; recommending the registration of an answer to the one or more questions; storing question and answer information indicating question and answer pairs, which are combinations of registered questions and answers; A question generation method including:

13. For the question and answer generation system, acquiring one or more pieces of dialogue information; generating one or more summaries from each of the dialogue information using a summary generation model; generating one or more questions from the plurality of summaries using a question generation model; recommending the registration of an answer to the one or more questions; storing question and answer information indicating question and answer pairs, which are combinations of registered questions and answers; A program for executing a question generation method including the steps of:

Images